Lecture 13: The Knapsack Problem Outline of this Lecture Introduction of the 0-1 Knapsack Problem. A dynamic programming solution to this problem. 1
0-1 Knapsack Problem Informal Description: We have computed data files that we want to store, and we have available bytes of storage. File has size bytes and takes minutes to recompute. We want to avoid as much recomputing as possible, so we want to find a subset of files to store such that The files have combined size at most. The total computing time of the stored files is as large as possible. We can not store parts of files, it is the whole file or nothing. How should we select the files? 2
0-1 Knapsack Problem Formal description: Given two -tuples of positive numbers and and, we wish to determine the subset!#" %$&'( *) (of files to store) that maximizes. subject to,+-,+- 0/ Remark: This is an optimization problem. Brute force: Try all $ possible subsets. Question: Any solution better than the brute-force? 3
Recall of the Divide-and-Conquer 1. Partition the problem into subproblems. 2. Solve the subproblems. 3. Combine the solutions to solve the original one. Remark: If the subproblems are not independent, i.e. subproblems share subsubproblems, then a divideand-conquer algorithm repeatedly solves the common subsubproblems. Thus, it does more work than necessary! Question: Any better solution? Yes Dynamic programming (DP)! 4
The Idea of Dynamic Programming Dynamic programming is a method for solving optimization problems. The idea: Compute the solutions to the subsub-problems once and store the solutions in a table, so that they can be reused (repeatedly) later. Remark: We trade space for time. 5
The Idea of Developing a DP Algorithm Step1: Structure: Characterize the structure of an optimal solution. Decompose the problem into smaller problems, and find a relation between the structure of the optimal solution of the original problem and the solutions of the smaller problems. Step2: Principle of Optimality: Recursively define the value of an optimal solution. Express the solution of the original problem in terms of optimal solutions for smaller problems. 6
The Idea of Developing a DP Algorithm Step 3: Bottom-up computation: Compute the value of an optimal solution in a bottom-up fashion by using a table structure. Step 4: Construction of optimal solution: Construct an optimal solution from computed information. Steps 3 and 4 may often be combined. 7
Remarks on the Dynamic Programming Approach Steps 1-3 form the basis of a dynamic-programming solution to a problem. Step 4 can be omitted if only the value of an optimal solution is required. 8
Developing a DP Algorithm for Knapsack Step 1: Decompose the problem into smaller problems. We construct an array 1 2 34 5 3 6. For " / /, and / /, the entry 1 27 8( 6 will store the maximum (combined) computing time of any subset of files!#" %$&( 9) of (combined) size at most. If we can compute all the entries of this array, then the array entry 1 27 5 6 will contain the maximum computing time of files that can fit into the storage, that is, the solution to our problem. 9
Developing a DP Algorithm for Knapsack Step 2: Recursively define the value of an optimal solution in terms of solutions to smaller problems. Initial Settings: Set 1 2 : 60; for / /, no item 1 2< 8 65; = > for?, illegal Recursive Step: Use 1 27 8( 65; @ ACBED 1 27 = " 6 ( (F 1 27 = " ( = 6.G for " / /, / /. 10
Correctness of the Method for Computing 1 27 8( 6 Lemma: For " / /, / /, 1 27 8( 6H; @ ACBED 1 27 = " : 6 F 1 27 = " = 6 G Proof: To compute 1 2< 8 6 we note that we have only two choices for file : Leave!#" file : The best we can do with files %$&( = " ) and storage limit 1 27 = " 8 6 is. Take file (only possible if I / ): Then we gain of computing time, but have spent bytes of our storage. The best we can do with remaining files!j" %$K = " ) and storage D = G is 1 27 = " = 6. Totally, we get L F 1 27 = " : = 6. Note that if, then F 1 27 = " 8 = 60; = > so the lemma is correct in any case. 11
Developing a DP Algorithm for Knapsack Step 3: Bottom-up computing 1 27 8 6 (using iteration, not recursion). Bottom: 1 2 6H; for all / /. Bottom-up computation: Computing the table using 1 2< 8 6H; @ AMBED 1 2< = " 6 : NF 1 2< = " = 6.G row by row. V[i,w] w=0 1 2 3...... W i= 0 1 2. n 0 0 0 0...... 0 bottom up 12
Example of the Bottom-up computation Let ; " and O V V 1 2 3 4 10 40 30 50 5 4 6 3 P,QSRUT XW 1 2 3 4 5 6 7 8 9 10 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 10 10 10 10 10 10 2 0 0 0 0 40 40 40 40 40 50 50 3 0 0 0 0 40 40 40 40 40 50 70 4 0 0 0 50 50 50 50 90 90 90 90 Remarks: Y The final output is O P[Z Q\V]TXW ^_V. Y The method described does not tell which subset gives the optimal solution. (It is ` CQ Zba in this example). 13
v The Dynamic Programming Algorithm KnapSack(c QSRdQ Qfe ) g (R W V e O P VCQhRiTW V for ( W ) ; for to (R W V ) e for (R P 3Tj to ) O P,QkRUTW R if ) l mon ` O P Mp qqhrutrq c P 3Tts O P Mp qqhr p R P 3T3T a ; else O P,QkRUTW OIP Mp qqhrut ; return O P Que T ; Time complexity: Clearly, w D G. 14
Constructing the Optimal Solution The algorithm for computing 1 27 8 6 described in the previous slide does not keep record of which subset of items gives the optimal solution. To compute the actual subset, we can add an auxiliary boolean array x#y]y(z*27 8{ 6 which is 1 if we decide to take the -th file in 1 2< 8 6 and 0 otherwise. Question: How do we use all the values x#y]y(z*2< 8 6 to determine the subset of files having the maximum computing time? 15
Constructing the Optimal Solution Question: How do we use the values x#yqy(z 27 8% 6 to determine the subset of items having the maximum computing time? If keep[ 5 } ] is 1, then. We can now repeat this argument for keep[ = " = ]. If keep[ H } ~ ] is 0, the and we repeat the argument for keep[ = " ]. Therefore, the following partial program will output the elements of : ; ; for ( ; downto 1) if (keep2< 8 60; ; " ) output ; i; = 27 6 ; 16
v v v The Complete Algorithm for the Knapsack Problem KnapSack(c QSRdQ Qfe ) g for (R W V to e ) O P VMQhRUTW V ; for ( W to ) for (R W V to e ) if ((R P 3TXj R ) and (c P 3Tts O P Cp qqhr p R P 3T3T O P Mp qqhrut )) g OIP,QhRUTW c P 3Tts O P Cp qqhr p R P 3T3T ; keep P,QƒRUTW ; g else OIP,QhRUTW O P Mp qqhrut v P,QƒRUTW V ; keep ; W e ( W ; for downto P,Q g TJW 1) W if (keep ) output W i; p R P 3T ; return O P Q e T ; 17